Add SplitUVPlanes and MergeUVPlanes

Add public methods SplitUVPlanes and MergeUVPlanes based on the
optimized assembly functions that already exists. Also, de-duplicate the
CPU dispatching code for these functions by moving them to helper
functions.

BUG=libyuv:629
R=braveyao@chromium.org

Review URL: https://codereview.chromium.org/2277603004 .

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1613
+Version: 1614
 License: BSD
 License File: LICENSE
 

include/libyuv/planar_functions.h

@@ -39,6 +39,20 @@ void SetPlane(uint8* dst_y, int dst_stride_y,
               int width, int height,
               uint32 value);
 
+// Split interleaved UV plane into separate U and V planes.
+LIBYUV_API
+void SplitUVPlane(const uint8* src_uv, int src_stride_uv,
+                  uint8* dst_u, int dst_stride_u,
+                  uint8* dst_v, int dst_stride_v,
+                  int width, int height);
+
+// Merge separate U and V planes into one interleaved UV plane.
+LIBYUV_API
+void MergeUVPlane(const uint8* src_u, int src_stride_u,
+                  const uint8* src_v, int src_stride_v,
+                  uint8* dst_uv, int dst_stride_uv,
+                  int width, int height);
+
 // Copy I400.  Supports inverting.
 LIBYUV_API
 int I400ToI400(const uint8* src_y, int src_stride_y,

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1613
+#define LIBYUV_VERSION 1614
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/convert.cc

@@ -40,14 +40,14 @@ static int I4xxToI420(const uint8* src_y, int src_stride_y,
   const int dst_y_height = Abs(src_y_height);
   const int dst_uv_width = SUBSAMPLE(dst_y_width, 1, 1);
   const int dst_uv_height = SUBSAMPLE(dst_y_height, 1, 1);
-  if (src_y_width == 0 || src_y_height == 0 ||
-      src_uv_width == 0 || src_uv_height == 0) {
+  if (src_uv_width == 0 || src_uv_height == 0) {
     return -1;
   }
-  // TODO(fbarchard): support NULL for dst_y
-  ScalePlane(src_y, src_stride_y, src_y_width, src_y_height,
-             dst_y, dst_stride_y, dst_y_width, dst_y_height,
-             kFilterBilinear);
+  if (dst_y) {
+    ScalePlane(src_y, src_stride_y, src_y_width, src_y_height,
+               dst_y, dst_stride_y, dst_y_width, dst_y_height,
+               kFilterBilinear);
+  }
   ScalePlane(src_u, src_stride_u, src_uv_width, src_uv_height,
              dst_u, dst_stride_u, dst_uv_width, dst_uv_height,
              kFilterBilinear);
@@ -229,75 +229,6 @@ static void CopyPlane2(const uint8* src, int src_stride_0, int src_stride_1,
   }
 }
 
-// Support function for NV12 etc UV channels.
-// Width and height are plane sizes (typically half pixel width).
-static void SplitUVPlane(const uint8* src_uv, int src_stride_uv,
-                         uint8* dst_u, int dst_stride_u,
-                         uint8* dst_v, int dst_stride_v,
-                         int width, int height) {
-  int y;
-  void (*SplitUVRow)(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
-                     int width) = SplitUVRow_C;
-  // Negative height means invert the image.
-  if (height < 0) {
-    height = -height;
-    dst_u = dst_u + (height - 1) * dst_stride_u;
-    dst_v = dst_v + (height - 1) * dst_stride_v;
-    dst_stride_u = -dst_stride_u;
-    dst_stride_v = -dst_stride_v;
-  }
-  // Coalesce rows.
-  if (src_stride_uv == width * 2 &&
-      dst_stride_u == width &&
-      dst_stride_v == width) {
-    width *= height;
-    height = 1;
-    src_stride_uv = dst_stride_u = dst_stride_v = 0;
-  }
-#if defined(HAS_SPLITUVROW_SSE2)
-  if (TestCpuFlag(kCpuHasSSE2)) {
-    SplitUVRow = SplitUVRow_Any_SSE2;
-    if (IS_ALIGNED(width, 16)) {
-      SplitUVRow = SplitUVRow_SSE2;
-    }
-  }
-#endif
-#if defined(HAS_SPLITUVROW_AVX2)
-  if (TestCpuFlag(kCpuHasAVX2)) {
-    SplitUVRow = SplitUVRow_Any_AVX2;
-    if (IS_ALIGNED(width, 32)) {
-      SplitUVRow = SplitUVRow_AVX2;
-    }
-  }
-#endif
-#if defined(HAS_SPLITUVROW_NEON)
-  if (TestCpuFlag(kCpuHasNEON)) {
-    SplitUVRow = SplitUVRow_Any_NEON;
-    if (IS_ALIGNED(width, 16)) {
-      SplitUVRow = SplitUVRow_NEON;
-    }
-  }
-#endif
-#if defined(HAS_SPLITUVROW_DSPR2)
-  if (TestCpuFlag(kCpuHasDSPR2) &&
-      IS_ALIGNED(dst_u, 4) && IS_ALIGNED(dst_stride_u, 4) &&
-      IS_ALIGNED(dst_v, 4) && IS_ALIGNED(dst_stride_v, 4)) {
-    SplitUVRow = SplitUVRow_Any_DSPR2;
-    if (IS_ALIGNED(width, 16)) {
-      SplitUVRow = SplitUVRow_DSPR2;
-    }
-  }
-#endif
-
-  for (y = 0; y < height; ++y) {
-    // Copy a row of UV.
-    SplitUVRow(src_uv, dst_u, dst_v, width);
-    dst_u += dst_stride_u;
-    dst_v += dst_stride_v;
-    src_uv += src_stride_uv;
-  }
-}
-
 // Support converting from FOURCC_M420
 // Useful for bandwidth constrained transports like USB 1.0 and 2.0 and for
 // easy conversion to I420.
@@ -324,7 +255,9 @@ static int X420ToI420(const uint8* src_y,
   if (height < 0) {
     height = -height;
     halfheight = (height + 1) >> 1;
-    dst_y = dst_y + (height - 1) * dst_stride_y;
+    if (dst_y) {
+      dst_y = dst_y + (height - 1) * dst_stride_y;
+    }
     dst_u = dst_u + (halfheight - 1) * dst_stride_u;
     dst_v = dst_v + (halfheight - 1) * dst_stride_v;
     dst_stride_y = -dst_stride_y;

source/convert_from.cc

@@ -361,6 +361,7 @@ int I420ToUYVY(const uint8* src_y, int src_stride_y,
   return 0;
 }
 
+// TODO(fbarchard): test negative height for invert.
 LIBYUV_API
 int I420ToNV12(const uint8* src_y, int src_stride_y,
                const uint8* src_u, int src_stride_u,
@@ -368,76 +369,19 @@ int I420ToNV12(const uint8* src_y, int src_stride_y,
                uint8* dst_y, int dst_stride_y,
                uint8* dst_uv, int dst_stride_uv,
                int width, int height) {
-  int y;
-  void (*MergeUVRow_)(const uint8* src_u, const uint8* src_v, uint8* dst_uv,
-      int width) = MergeUVRow_C;
-  // Coalesce rows.
-  int halfwidth = (width + 1) >> 1;
-  int halfheight = (height + 1) >> 1;
-  if (!src_u || !src_v || !dst_uv ||
+  if (!src_y || !src_u || !src_v || !dst_y || !dst_uv ||
       width <= 0 || height == 0) {
     return -1;
   }
-  // Negative height means invert the image.
-  if (height < 0) {
-    height = -height;
-    halfheight = (height + 1) >> 1;
-    if (dst_y) {
-      dst_y = dst_y + (height - 1) * dst_stride_y;
-    }
-    dst_uv = dst_uv + (halfheight - 1) * dst_stride_uv;
-    dst_stride_y = -dst_stride_y;
-    dst_stride_uv = -dst_stride_uv;
-  }
-  if (src_stride_y == width &&
-      dst_stride_y == width) {
-    width *= height;
-    height = 1;
-    src_stride_y = dst_stride_y = 0;
-  }
-  // Coalesce rows.
-  if (src_stride_u == halfwidth &&
-      src_stride_v == halfwidth &&
-      dst_stride_uv == halfwidth * 2) {
-    halfwidth *= halfheight;
-    halfheight = 1;
-    src_stride_u = src_stride_v = dst_stride_uv = 0;
-  }
-#if defined(HAS_MERGEUVROW_SSE2)
-  if (TestCpuFlag(kCpuHasSSE2)) {
-    MergeUVRow_ = MergeUVRow_Any_SSE2;
-    if (IS_ALIGNED(halfwidth, 16)) {
-      MergeUVRow_ = MergeUVRow_SSE2;
-    }
-  }
-#endif
-#if defined(HAS_MERGEUVROW_AVX2)
-  if (TestCpuFlag(kCpuHasAVX2)) {
-    MergeUVRow_ = MergeUVRow_Any_AVX2;
-    if (IS_ALIGNED(halfwidth, 32)) {
-      MergeUVRow_ = MergeUVRow_AVX2;
-    }
-  }
-#endif
-#if defined(HAS_MERGEUVROW_NEON)
-  if (TestCpuFlag(kCpuHasNEON)) {
-    MergeUVRow_ = MergeUVRow_Any_NEON;
-    if (IS_ALIGNED(halfwidth, 16)) {
-      MergeUVRow_ = MergeUVRow_NEON;
-    }
-  }
-#endif
-
+  int halfwidth = (width + 1) / 2;
+  int halfheight = height > 0 ? (height + 1) / 2 : (height - 1) / 2;
   if (dst_y) {
     CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
   }
-  for (y = 0; y < halfheight; ++y) {
-    // Merge a row of U and V into a row of UV.
-    MergeUVRow_(src_u, src_v, dst_uv, halfwidth);
-    src_u += src_stride_u;
-    src_v += src_stride_v;
-    dst_uv += dst_stride_uv;
-  }
+  MergeUVPlane(src_u, src_stride_u,
+               src_v, src_stride_v,
+               dst_uv, dst_stride_uv,
+               halfwidth, halfheight);
   return 0;
 }
 

source/planar_functions.cc

@@ -31,6 +31,12 @@ void CopyPlane(const uint8* src_y, int src_stride_y,
                int width, int height) {
   int y;
   void (*CopyRow)(const uint8* src, uint8* dst, int width) = CopyRow_C;
+  // Negative height means invert the image.
+  if (height < 0) {
+    height = -height;
+    dst_y = dst_y + (height - 1) * dst_stride_y;
+    dst_stride_y = -dst_stride_y;
+  }
   // Coalesce rows.
   if (src_stride_y == width &&
       dst_stride_y == width) {
@@ -76,6 +82,7 @@ void CopyPlane(const uint8* src_y, int src_stride_y,
   }
 }
 
+// TODO(fbarchard): Consider support for negative height.
 LIBYUV_API
 void CopyPlane_16(const uint16* src_y, int src_stride_y,
                   uint16* dst_y, int dst_stride_y,
@@ -224,6 +231,133 @@ int I420ToI400(const uint8* src_y, int src_stride_y,
   return 0;
 }
 
+// Support function for NV12 etc UV channels.
+// Width and height are plane sizes (typically half pixel width).
+LIBYUV_API
+void SplitUVPlane(const uint8* src_uv, int src_stride_uv,
+                  uint8* dst_u, int dst_stride_u,
+                  uint8* dst_v, int dst_stride_v,
+                  int width, int height) {
+  int y;
+  void (*SplitUVRow)(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
+                     int width) = SplitUVRow_C;
+  // Negative height means invert the image.
+  if (height < 0) {
+    height = -height;
+    dst_u = dst_u + (height - 1) * dst_stride_u;
+    dst_v = dst_v + (height - 1) * dst_stride_v;
+    dst_stride_u = -dst_stride_u;
+    dst_stride_v = -dst_stride_v;
+  }
+  // Coalesce rows.
+  if (src_stride_uv == width * 2 &&
+      dst_stride_u == width &&
+      dst_stride_v == width) {
+    width *= height;
+    height = 1;
+    src_stride_uv = dst_stride_u = dst_stride_v = 0;
+  }
+#if defined(HAS_SPLITUVROW_SSE2)
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    SplitUVRow = SplitUVRow_Any_SSE2;
+    if (IS_ALIGNED(width, 16)) {
+      SplitUVRow = SplitUVRow_SSE2;
+    }
+  }
+#endif
+#if defined(HAS_SPLITUVROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    SplitUVRow = SplitUVRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
+      SplitUVRow = SplitUVRow_AVX2;
+    }
+  }
+#endif
+#if defined(HAS_SPLITUVROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON)) {
+    SplitUVRow = SplitUVRow_Any_NEON;
+    if (IS_ALIGNED(width, 16)) {
+      SplitUVRow = SplitUVRow_NEON;
+    }
+  }
+#endif
+#if defined(HAS_SPLITUVROW_DSPR2)
+  if (TestCpuFlag(kCpuHasDSPR2) &&
+      IS_ALIGNED(dst_u, 4) && IS_ALIGNED(dst_stride_u, 4) &&
+      IS_ALIGNED(dst_v, 4) && IS_ALIGNED(dst_stride_v, 4)) {
+    SplitUVRow = SplitUVRow_Any_DSPR2;
+    if (IS_ALIGNED(width, 16)) {
+      SplitUVRow = SplitUVRow_DSPR2;
+    }
+  }
+#endif
+
+  for (y = 0; y < height; ++y) {
+    // Copy a row of UV.
+    SplitUVRow(src_uv, dst_u, dst_v, width);
+    dst_u += dst_stride_u;
+    dst_v += dst_stride_v;
+    src_uv += src_stride_uv;
+  }
+}
+
+LIBYUV_API
+void MergeUVPlane(const uint8* src_u, int src_stride_u,
+                  const uint8* src_v, int src_stride_v,
+                  uint8* dst_uv, int dst_stride_uv,
+                  int width, int height) {
+  int y;
+  void (*MergeUVRow)(const uint8* src_u, const uint8* src_v, uint8* dst_uv,
+      int width) = MergeUVRow_C;
+  // Coalesce rows.
+  // Negative height means invert the image.
+  if (height < 0) {
+    height = -height;
+    dst_uv = dst_uv + (height - 1) * dst_stride_uv;
+    dst_stride_uv = -dst_stride_uv;
+  }
+  // Coalesce rows.
+  if (src_stride_u == width &&
+      src_stride_v == width &&
+      dst_stride_uv == width * 2) {
+    width *= height;
+    height = 1;
+    src_stride_u = src_stride_v = dst_stride_uv = 0;
+  }
+#if defined(HAS_MERGEUVROW_SSE2)
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    MergeUVRow = MergeUVRow_Any_SSE2;
+    if (IS_ALIGNED(width, 16)) {
+      MergeUVRow = MergeUVRow_SSE2;
+    }
+  }
+#endif
+#if defined(HAS_MERGEUVROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    MergeUVRow = MergeUVRow_Any_AVX2;
+    if (IS_ALIGNED(width, 32)) {
+      MergeUVRow = MergeUVRow_AVX2;
+    }
+  }
+#endif
+#if defined(HAS_MERGEUVROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON)) {
+    MergeUVRow = MergeUVRow_Any_NEON;
+    if (IS_ALIGNED(width, 16)) {
+      MergeUVRow = MergeUVRow_NEON;
+    }
+  }
+#endif
+
+  for (y = 0; y < height; ++y) {
+    // Merge a row of U and V into a row of UV.
+    MergeUVRow(src_u, src_v, dst_uv, width);
+    src_u += src_stride_u;
+    src_v += src_stride_v;
+    dst_uv += dst_stride_uv;
+  }
+}
+
 // Mirror a plane of data.
 void MirrorPlane(const uint8* src_y, int src_stride_y,
                  uint8* dst_y, int dst_stride_y,